AI Programming Lecture 8 Natural Language Processing Semantics

1
AI ProgrammingLecture 8Natural Language
ProcessingSemantics

• Richard Price Simon Worgan
• School Of Computer Science University Of
  Birmingham
• msc59rmp, msc74sxw_at_cs.bham.ac.uk

2
Last Week

• NLP
• Applications.
• Problems.
• Syntax.
• Parsers.
• Recognisers.

3
Introduction

• The importance of syntax.
• Blocksworld.
• Syntax Semantics.
• Simple understanding responses.
• Pragmatics.
• Eliza.
• NLP Mini-Project.

4
Compositionality

• Syntactic structure is useful.
• Structure often indicates meaning.
• Freges Principle
• The meaning of a complex expression is determined
  by the meanings of its components.

5
Blocksworld Environment

• Environment status.
• isa block objR
• colour red objR
• size big objR
• objR ison table
• include objG
• ? database.

6
Which

• Built in database query tool.
• which().
• lvars v
• which(v, size big ?visa block ?v) gt
• objR
• Which(v, is a block ?v) gt
• objG objR

7
Blocksworld Grammar

• Sample grammar with lexicon.
• s wh1 vbe np
• np det snp
• snp n
• snp ap n
• ap adj
• wh1 where
• det the
• n box
• vbe is
• ? database.

8
Top Down Parser

• For actual code see Sheilas notes.
• backwardsParseTree()
• Start with high level goal structure e.g S.
• Taking the rules in order
• If the LHS of a rule matches our goal or
  sub-goal.
• Then explore the RHS of this rule as a sub-goal.
• Until either
• We complete all our goals sub-goals.
• We run out of rules.
• Return structure.

9
Parsing Example
S
Where is the red block
10
Meaning

• Search the syntax tree for words of interest.
• Returns potential database queries.
• meaning(n block) gt
• isa block ? x
• meaning( s wh1 where vbe is np det the
  snp n block ) gt
• where isa block ? x
• Which return information on the environment.

11
Intelligent Responses

• The information leads to more queries.
• Gradually being more specific.
• isa block ? x
• which(x, isa block ? x) gt
• objG objR
• Which can then be relayed to the user.
• If ?x ison ?y isin database then
• x is on top of ?y gt

12
Pragmatics

• Connect language to the world.
• Consider intention and context.
• Conversational background
• Implicit beliefs shared by the participants.

13
Context - Semantics

• The unspoken sentences

14
Anaphora

• Context persists across time
• Ten undergraduates and four postgraduate
  linguists attended the seminar. They were
  constantly heckling them.
• They? Them?
• The postgraduate and undergraduate groups persist
  over time.
• OK i've got the nail. Now, when I nod my head
  you hit it.
• It?

15
Intention

• Conversations convey meanings.
• Therefore we express meanings, using words, in
  conversation.
• Which words?
• We reason about the the beliefs and goals of the
  listener.

16
Grounding

• Webster's Dictionary

• Where is the actual meaning?

17
Breaking Eliza
You HelloEliza WHY ARE YOU SO
UNFORTHCOMING? You I'm just saying helloEliza
COULD IT BE THAT YOU ARE TOO SELF-CENTRED? You
No not really, thats a little harsh you've only
just met meEliza WHY NOT REALLY THATS A LITTLE
HARSH ME HAVE ONLY JUST MET YOU

• Eh?
• Only the Syntax is considered

18
Possible Expansions to Eliza

• Anaphoric references You've Me
• Conversational background
• The start of a conversation.
• Levels of familiarity.
• Intention
• Defensive.
• Conveying a feeling of being insulted.
• Meaning?
• Tricky

19
MP Why communicate?

• Agent motivation
• Inform, Query, Answer, Request, Promise,
  Acknowledge, Share.
• Is this useful?
• What about lying?

20
MP Natural Language Ideas

• 'Classic' NLP systems.
• Ambiguity
• Users motivation/intention.
• Situated NLP, Agent Systems.
• Environment -gt Pragmatics
• Intention
• Converstations

21
Conclusion

• Syntax is a precursor to Semantics.
• Simple NL database query tool.
• Advanced Semantics requires more
• Pragmatics
• Grounding
• How do we do it?